Devin (Cognition)
Structured Knowledge items stored server-side. Three-tier scope: repo-pinned (always injected), global (recall by relevance), org/enterprise (visible to org members). Auto-generates Repo Knowledge from repo structure each session; reads past session trajectories for self-improvement; writes notes for recurring scheduled tasks.
At a glance
- Type
- Server-side Knowledge + auto-Repo Knowledge
- Tier
- T1
- Section
- Coding-agent memory
- Created
- 2024-03
- Latest release
- not applicable — not OSS
- License
- not applicable — not OSS
- GitHub
- not applicable — no GitHub repo
- Pricing
- Free + paid
- Funding
- $1.1B total $10.2B val Series E · 2025-09
Taxonomy
- storage
- kv
- retrieval
- injection
- persistence
- cross-session
- update
- agent-controlled
- unit
- fact
- governance
- inspectable
- conflict
- none
When to use
Optimised for: autonomous SWE work + structured Knowledge
Anti-fit: not for non-coding tasks; high cost per session
Pros & cons
Pros
Most aggressive memory model in the coding-agent space — explicit knowledge base + file-backed task notes + persistent terminal sessions.
Cons
Memory features only available inside the Devin product; not portable to other agents.
Claims & capabilities
$175M Series B at ~$2B (2024). Devin 2.0 announced Mar 2025.
Technical surface
- API surface
- REST, Slack integration
- Backend storage
- custom
- Deployment
- Both
- Embedding model
- locked
- Multi-tenancy
- Enterprise dedicated SaaS keeps customer data in customer tenant; on-prem and VPC deployment options available
- MCP
- via official adapter — Devin supports MCP servers
- A2A
- searched not found
- OpenTelemetry
- searched not found
Compare Devin (Cognition) with…
Similar systems
Other coding-agent memory in the catalog, ranked by inbound references.
- Aider T2
Appends conversation turns to .aider.chat.history.md in project root. restore_chat_history defaults to false — no automatic replay into new sessions. Community /session save command for JSON state snapshots. .aider.conf.yml carries static preferences.
- Augment Code T1
Real-time knowledge graph indexed from connected repos + commit history + CI/CD + tickets + docs. Separate Agent Memory layer (Sept 2025) captures session decisions/corrections/goals as discrete items with developer review/approval UI. MCP endpoint exposes the graph.
- Bolt.new (StackBlitz) T2
In-browser WebContainer (Rust-backed shared memory) is the implicit state store. No dedicated memory abstraction — project files are the state. Team plan adds shared workspaces + GitHub integration.
- Continue.dev Memory MCP T2
Official Memory MCP (Mar 2025) backed by a knowledge graph. Memory is injected via MCP rather than native to the Continue core. Scope depends on MCP server deployment (local or shared).
- Lovable T2
Long-term context stored as files in /docs that the agent reads on each session. Agent Mode (default since July 2025) holds working context within session but doesn't persist cross-session beyond filesystem state.
- Mistral Vibe (Remote Agents) + Mistral Medium 3.5 T1
Vibe is Mistral's cloud-hosted coding-agent platform: each session runs in an isolated sandbox with task state, history, and approvals. Spawnable from CLI or Le Chat, with parallel execution. Local CLI sessions can be "teleported" to the cloud and resumed. Powered by Mistral Medium 3.5 (77.6% SWE-Bench Verified). Plugs into GitHub, Linear/Jira, Sentry, Slack/Teams.